One of the most widely used frameworks in Silicon Valley is “Innovator’s Dilemma”, popularized by the late Clayton Christensen. In this post, I’ll argue why that framework fails for enterprise AI adoption, which is unfortunate for most AI startups.
As a reminder, Innovator’s Dilemma basically says that incumbents can lose to startups by either being late to - or ignoring - small but fast growing markets that are created by technological innovations or new distribution channels. Incumbents make these mistakes of inaction - not because they are stupid or technologically inferior to startups - but because they are slowed down by internal politics, worries over cannibalization, wishful thinking, etc. And by the time it’s too late, the upstarts have built significant distribution and own the new category.
But unfortunately for startups, Innovator’s Dilemma doesn’t really apply to the enterprise AI space, because:
the incumbents and startups both want the same thing, at the same time, hence there’s no dilemma: both incumbents (from FAANG to mid-market SIs) and startups are going for the same markets, the same customers, at the same time, with fairly undifferentiated approaches. This is painfully obvious if you sit on both sides of the table - pitching or getting pitched. This makes sales motions very difficult, whether you are selling a vertical or horizontal solutions.
startups don’t have speed advantage, because sales cycles are long enough where incumbents can catch up on the “innovation”: since enterprise sales cycles are long, startups have no “speed advantage” versus Google or AWS, since by the time startups are actually “production-ready”, there will be a competing offering from Big Companies. Case in point, Langchain’s Langfuse versus Google’s GenKit / LLM ops solutions.
incumbents have more information than startups about the customer demand: the incumbents are undeterred to compete directly with their integrators or ISVs wherever there’s traction, and they know more about what’s working than any individual ISV or startups do. And unlike pre-2022, these incumbents are actually serious about shipping and iterating based on real demand signals from real customer conversations.
boom and bust cycles are too short for startups to have any sustainable advantage: with each Google I/O or OpenAI announcement, at least a few dozen well-funded AI startups are affected, since everyone’s innovating on similar axis’s of value. There’s very few orthogonal takes on the “AI idea space” that is decoupled from Google or OpenAI’s product roadmaps. You really need to think of something that the 1000+ Google PMs can’t collectively think of, and that’s hard. And even if you do, your idea will get stolen once you announce it on Twitter and get any traction.
there’s no new distribution channel for startups to tap: another way Innovator’s Dilemma could work is when startups take advantage of distribution channels that the incumbents are ignoring. But in 2024, the channels to sell enterprise are very well established and crowded. You don’t get easy “organic buzz” just by being a YC backed company like you used to in 2014, anymore. And even then, it was hard to land meetings with enterprise.
These observations apply whether you are an AI startup selling a horizontal or vertical solution, selling to large enterprise to market, or a software pureplay or service enhanced. To keep this post short, I’ll discuss why in a future post, as the conclusions are counter intuitive.
Of course, there are some exceptions like Perplexity that raised a seed round from Elad Gil and got backed by Jeff Bezos early. These are the companies that have found a good wedge to enter a big market, but more importantly, found the right investors to become relevant - even if they are basically ChatGPT wrappers. With enough endorsements and funding, you will be relevant as long as your product’s decent.
If you are not Perplexity, startups should be careful about their GTM motions, and especially who they are trying to sell to - to avoid inadvertently waking up the incumbents. But that’s pretty difficult when you are selling any form of software that depends on the Internet as a scalable sales channel.